By asahifile 
For most of the internet’s life there were two kinds of visitor: people and bots. Sites tolerated bots for narrow automated jobs and wrote rules for them, because the web was built for people first. Personal AI agents have broken that division, and the web has not decided what to do about it.
As MIT Technology Review reported this week, agents acting on behalf of individual users — booking flights, reserving hotels, hunting for a better price — are arriving at scale through products such as OpenAI’s Dots, Meta’s Muse, Grok and Instinct. Muse alone was downloaded five million times in its first month, and agentic web traffic tripled between August and September.
Why the old rules do not fit
Anti-bot systems were designed around a simple assumption: if a request is not from a human, it is suspect. Agents break that test. They are not people, and they operate at machine speed, but they are not scrapers either — they act for a named individual, on legitimate business, faster than that individual could. Faced with a visitor that is neither, many platforms have simply banned it.
The blocking is already broad. Yelp and Delta have blocked agent traffic, Amazon blocked Muse’s agents and is litigating against Perplexity over its agents, and security-conscious financial firms including JPMorgan Chase and Charles Schwab have shut out AI access. Meanwhile Shopify has gone the other way, giving agents their own checkout lanes that skip the pages designed for people.
The direction of travel is not in doubt
Shopify’s choice points to the answer. Most agentic traffic heads for e-commerce, which means there is money in letting agents in, and money usually wins. Once agents can shop, compare and negotiate at volume, sites will start optimising for the visitor that generates more requests — and that visitor will not be you.
The consequences are easy to sketch. Human-facing interfaces get neglected because they serve the minority. Verification systems get cranked up until real people are constantly proving they are not machines. Prices stop being a single number for everyone and start varying by whose agent is asking. And the recommendation your agent gives you may or may not be shaped by an advertisement or a commission you cannot see.
Wharton’s Ethan Mollick has made a sharper version of the same point: once it is well known that agents can be sent to request refunds or haggle over prices, customer service desks will be flooded to the point of being unusable for actual humans.
The case for letting them in
It would be wrong to read this as an argument for banning agents. They can do genuinely useful work: contesting a bogus medical bill, comparing car insurance, chasing a refund the customer is owed but would never pursue. Those are exactly the chores automation should absorb. The question is not whether agents belong on the web; it is what rules govern them, and who writes those rules.
Some of the answers are easy and uncontroversial — nobody wants card details waved around by roaming agents, and stronger agent authentication is plainly good. The contested ones are about access and leverage: may a business block an agent a customer deliberately sent to negotiate on their behalf, and must an agent disclose when it is acting for the platform rather than the person? Those decisions will settle whether the agent web works for users or against them — and they are being made now, mostly in private.
What to watch
- The outcome of Amazon’s litigation with Perplexity, which will shape whether blocking agents is legally straightforward.
- Whether standards bodies and browser vendors produce an authentication scheme that lets sites tell a user’s agent from a scraper.
- Whether “agent checkout” lanes like Shopify’s spread, and whether human checkout quietly degrades behind them.
- Whether disclosure rules require agents to reveal when a recommendation is commercially influenced.
Sources: MIT Technology Review and aggregated reporting on agent blocking via Ground News.